The research provides a theoretical basis for the idea that the brain works near a limit state (critical in the professional language), and this is important both for the characterization of the dynamics of the healthy brain and for the understanding of possible mechanisms for the development of disorders
A recent study conducted by Dr. Oren Shariki from the Department of Psychology and Neuroscience and Cognition at Ben-Gurion University of the Negev, in collaboration with Dobi Yelin from the Weizmann Institute of Science, suggests that learning processes in the brain can lead networks of nerve cells in the brain to operate in a unique borderline environment. On the one hand, working in this mode allows neural networks to perform optimally in the ability to process sensory information. On the other hand, this also involves the risk of "crossing the line" to a situation where the network activity does not represent the input and hallucinations develop.
On the applied level, the research provides a theoretical basis for the idea that the brain works near a limit state (critical in the professional language), and this is important both for the characterization of the dynamics of the healthy brain and for the understanding of possible mechanisms for the development of disorders. The study was published on February 16, 2016 in the journal PLoS Computational Biology, which allows open access to its articles.
It should be noted that neural networks in the brain include feedforward connections, which transfer information from one processing stage to another. However, neurons in each processing stage "talk" to each other through recurrent feedback connections. Feedback networks can exhibit a rich repertoire of dynamic behaviors. In particular, they can maintain neural activity even without objective external input, a situation that can be interpreted as a hallucination. A central question is how the feedback connections contribute to the representation of information in the brain and how they must develop to bring about optimal information representation. In order to answer this question, the researchers developed and studied a mathematical model of a neural network, which includes feed-forward connections from one layer of nerve cells to another layer as well as feedback connections between the neurons in the second layer. (from the chart)
Similar to real neural networks in the brain, the network model is presented with sensory inputs, and the feedback connections develop in the process of learning. The rules of plasticity that are responsible for changes in feedback connections are structured so that the representation of information in the network will improve during learning. The researchers found that optimal information representation is obtained when the network tends to operate close to the state of "hallucinations".
Dr. Shariki: This can be explained using a simple analogy: imagine that you have to speak through a microphone to a large audience. When you increase the volume on the amplifier, at some point you start to hear a sharp sound, resulting from the feedback between the microphone and the amplifier. To avoid this sound, which does not convey information, but still maintain a high volume, you weaken the volume a little. Similarly, optimal sensitivity to external sensory inputs is achieved when the feedback connections in the neural network are strong enough to amplify the inputs, but are not too dominant for hallucinations to occur."
These findings also provide insights into the possible causes of some neurological and psychiatric disorders. In particular, networks near the boundary condition are very sensitive to small changes in the strength of feedback connections between neurons. Such small changes can transfer a normal neural network to the unwanted state of hallucinations and manifest as a neurological disorder. For example, it is known that when people are in sensory isolation, they begin to develop hallucinations. The model offers an interesting explanation for the mechanism behind the phenomenon: when the inputs to the network are weakened, the network approaches the limit point in order to increase the inputs even more, but may cross the limit and develop hallucinations. According to Dr. Shariki, hallucinations are generally undesirable, as they indicate that the areas that process sensory information have become disconnected from their input. Our claim is that a state that is close to hallucinating, but without yet entering hallucinating, is computationally optimal. In terms of the connection to drugs, the claim that emerges from the article is that you don't need a large amount of a hallucinogenic drug to cause hallucinations, because the system is already close to the state of hallucinations, but it is difficult to quantify this.
Photo Caption: Dr. Oren Shariki.
http://imagelibrary.bgu.ac.il/pf.tlx/jRjLWjpSE2Q
12 תגובות
Feedback or feedback
relevant xkcd:
https://xkcd.com/323/
It's called mescaline, it's the only way to fly.
rival
As far as I understand - feedback in Hebrew means "feed back", or "feedback".
from neglecting the study of dreams,
It seems to me that the study misses a good opportunity to gather information
Regarding the way the brain works.
It is likely that the act of dreaming
It is an important part of the way the brain works in a state of alertness...
The article is incomprehensible and unclear and it seems as if the writer sends the reader to Wikipedia for every second sentence in which he uses a medical term that he does not bother to explain. Unintelligible and condescending. The lecturer is considered a bad lecturer, so he explains things that way
The article here is mathematical. The state that slides into instability is mathematical. Ray Kurzweil didn't do things in mathematics, he was a predictor of the future, a thinker.
They developed a mathematical model of the thought process in a neuron. They may have been inspired, but this mathematical model looks quite interesting.
Now the question arises as to whether anyone has built a mathematical model before. I think so. I seen. I am not from the field and cannot appreciate how innovative the work is.
You can listen to the segment in question (or download it as a file) starting at minute 02:10, here:
http://media.afb.org/2104/TechPioneers/Kurzweil-tape1sideAcont.mp3
Correction, feedback-back is of course Feed-Back.
"I do my most creative work, literally, sort of in bed, a lucid dreaming process. It's actually a mental discipline that I've done for a long time.
Before I go to sleep I assign myself some problem and it can be any kind of problem. It can be a business challenge, a decision or some business issue. It could be a math problem. It can be how to design a particular invention or it could be a writing problem: how do I organize material to express a certain thought.
And I try not to solve the problem, because obviously that will cut off any further creative thought about it. But I do try to reflect on what to do. I know about this? What would a solution look like? What would a solution look like and what's the background knowledge I have related to this.
And then I go to sleep and then I will dream about it and if I wake up in the middle of the night I will find myself dreaming about something having to do, in a strange way, with this problem. And there's a couple of interesting characteristics of our dream thinking process. Censors are relaxed in dreaming. We allow ourselves to think about things we don't allow ourselves to think about. That's why dreams very often will deal with social taboos and things we don't allow ourselves to think about or at least talk about during the day.
But also, there are professional taboos that are relaxed. Taboos like, "You can't solve a signal processing problem that way," and "you don't use that technique for linguistics," and we have all these ways that professionals learn that you are supposed to solve problems and those are relaxed as well so you start to think in creative new ways.
Something else not functioning in your dream life is your rational faculty. So you really are unable to evaluate ideas at that time. That's one of the reasons why strange things will happen in your dreams and what's strange about those, like say an elephant will come through the wall, what's strange about it is not just that the elephant came through the wall but the fact that you don' I don't think it's strange in the dream.
It happens and you don't say, "My God, the elephant came through the wall. How did that happen?" [You say,] “Okay, the elephant came through the wall. Nothing strange about it.”
It's because our rational faculties are not working. What you really would like to do is have this ability to make new connections, a sort of relaxation of the professional restrictions on our thinking, but also have our rational faculties.
Well there's a stage of in between dreaming and being awake. Kind of a lucid dreaming period. It generally happens in the morning. You are kind of vaguely aware that you're in bed. You could, for example, formulate the rational thought, "Gee, I've got a lot to do today. I'd better get out of bed and get up. That is, in fact, what most people do at that time.
But if you stay in bed instead, and let the dream process continue, cause you still have sort of access to the dream process, but now you're sufficiently conscious that your rational critical faculties are also working.
And I'll come back to the problem I assigned myself when I went to sleep and invariably I'll have really dramatic new insights into that problem. Very often a solution, new ways of thinking about it.
If I have an important decision that I'm really not sure about, you know, what strategy should I use for this business problem, or should we hire this person or do this business deal or how would I solve this technical problem, I will use this technique and generally that's how I will solve these few problems.
During the day I'm just sort of carrying out these dream decisions. I don't rush out of bed and I don't work twelve-hour days, but I try to be efficient. My mind is always thinking about these things so depending on what you mean by work, I am constantly thinking and my work is mental work. I'm dealing with ideas, whether I'm doing technology development, business development, or I'm doing these communications projects in terms of writing. It's all the world of ideas"
http://www.afb.org/info/living-with-vision-loss/using-technology/ray-kurzweil/part-2-of-4/1245
First, neural networks that include feed-forward content are not new:
https://www.youtube.com/results?search_query=recurrent+neural+networks
Secondly, the article reminded me of Ray Kurzweil, he claims that most of his ideas and inventions come from a lucid dream, which is actually a very realistic type of hallucination, which is quite consistent with what is written here.
The article here and in the original are fascinating. I cannot appreciate how innovative they are in relation to the world. The original article is presented very scientifically with machine learning models, differential equations and their solution. Identification of a threshold solution. It fascinates me how successful they are with a mathematical model to describe such things.